---

name: rnn-language-modeling description: "Plan and troubleshoot DARTS recurrent language-modeling workflows for PTB and WikiText-2." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

RNN Language Modeling

Use this sub-skill when a task is about DARTS recurrent language modeling on Penn Treebank or WikiText-2: architecture search, training from a recurrent genotype, test perplexity evaluation, corpus layout, checkpoint behavior, optimizer schedule behavior, or RNN-specific failures.

Routing

• Read references/workflows.md to plan PTB search, PTB train/test, the WikiText-2 recipe, major flags/defaults, checkpoint files, ASGD switching, rollback, and expected training signals.
• Read references/data-formats.md to validate train.txt / valid.txt / test.txt, <eos> tokenization, Dictionary / Corpus behavior, and batchify / get_batch tensor shapes.
• Read references/api-reference.md for RNNModel, DARTSCell, RNNModelSearch, Architect, dropout utilities, and checkpoint utilities.
• Read references/troubleshooting.md for dataset assertions, the inverted CUDA flag, CPU-only caveats, checkpoint loading, NaN rollback, hidden-size assertions, batch/micro-batch mismatches, and memory pressure.

Route genotype catalog lookup, recurrent DOT rendering, and visualization details to the sibling genotypes-and-visualization sub-skill. Route global legacy runtime compatibility and pretrained checkpoint acquisition caveats to the root skill's shared references.